This is an implementation for our TheWebConf 2023 paper: xGCN: An Extreme Graph Convolutional Network for Large-scale Social Link Prediction.

The current code is our original raw scripts. We release a refactored new version with improved functionality and detailed documentation: https://xgcn.readthedocs.io/en/latest/index.html

We provide examples and scripts to run the models: https://xgcn.readthedocs.io/en/latest/user_guide/usage_examples/example_pokec_livejournal.html

The Python version is 3.8.12, other dependencies are listed in requirements.txt.

• The .sh scripts to process datasets and run models are in the script directory.
• You are expected to specify two paths: (1) PROJECT_ROOT, which is the root of this project (e.g. /home/xxx/code/xgcn); (2) ALL_DATA_ROOT, which is going to hold all the data including datasets and models' outputs (e.g. /home/xxx/data/xgcn).
• The processed Pokec and LiveJournal datasets can be downloaded from here: https://drive.google.com/file/d/1lv-JkPmswYQ-YfKfRpNpoVh_N2XCiRQg/view?usp=drive_link. The original datasets are from http://snap.stanford.edu/data/index.html.
• The data (.txt files) should be placed in the directory $ALL_DATA_ROOT'/datasets/instance_[DATASET_NAME]' as follows (e.g. the pokec dataset):

instance_pokec
├── train.txt       # graph for training
├── validation.txt  # validation set
├── test.txt        # test set

• Dataset format
  • Each dataset contains 3 files: train.txt, validation.txt and test.txt.
  • Node ids start from 0, and are continuous integers.
  • train.txt contains the directed edges for training. Each line represents an edge: src dst.
  • validation.txt contains the edges for validation. Each line represents an edge: src dst.
  • test.txt contains the edges for test. Each line represents one edge or mutiple edges: src dst1 dst2 ....
• To process the raw .txt files, cd script and run the script process_data.sh (bash prcess_data.sh e.g. the dataset is pokec). After doing so, some new files for caching will be saved in directory instance_[DATASET_NAME]:

instance_pokec
├── train.txt
├── validation.txt
├── test.txt
├── info.yaml
├── train_csr_indptr.pkl
├── train_csr_indices.pkl
├── train_csr_src_indices.pkl
├── train_undi_csr_indptr.pkl
├── train_undi_csr_indices.pkl
├── train_undi_csr_src_indices.pkl
├── validation.pkl
├── test.pkl

• All the training settings and hyper-parameter configurations are in config and script/run_model (The cmd arguments in .sh files will overwrite those in .yaml config files).
• To run a model, modfiy PROJECT_ROOT, ALL_DATA_ROOT, DEVICE, DATASET and MODEL in script/run_model.sh. cd script and bash run_model.sh. The outputs (including training log, test results, ...) will be saved in $ALL_DATA_ROOT'/model_outputs/gnn_'$DATASET'/'$MODEL.